Tachyon: Reliable File Sharing at Memory- Speed Across Cluster Frameworks



Similar documents
Tachyon: memory-speed data sharing

Tachyon: A Reliable Memory Centric Storage for Big Data Analytics

An Open Source Memory-Centric Distributed Storage System

Spark: Making Big Data Interactive & Real-Time

Spark and Shark. High- Speed In- Memory Analytics over Hadoop and Hive Data

Mesos: A Platform for Fine- Grained Resource Sharing in Data Centers (II)

Spark in Action. Fast Big Data Analytics using Scala. Matei Zaharia. project.org. University of California, Berkeley UC BERKELEY

Fast and Expressive Big Data Analytics with Python. Matei Zaharia UC BERKELEY

GraySort on Apache Spark by Databricks

Spark. Fast, Interactive, Language- Integrated Cluster Computing

In-memory data pipeline and warehouse at scale using Spark, Spark SQL, Tachyon and Parquet

Beyond Hadoop with Apache Spark and BDAS

CSE-E5430 Scalable Cloud Computing Lecture 11

Hadoop Architecture. Part 1

Unified Big Data Processing with Apache Spark. Matei

Near Real Time Indexing Kafka Message to Apache Blur using Spark Streaming. by Dibyendu Bhattacharya

Conquering Big Data with BDAS (Berkeley Data Analytics)

Rakam: Distributed Analytics API

Introduction to Big Data! with Apache Spark" UC#BERKELEY#

Distributed File System. MCSN N. Tonellotto Complements of Distributed Enabling Platforms

Next-Gen Big Data Analytics using the Spark stack

Dell In-Memory Appliance for Cloudera Enterprise

How Companies are! Using Spark

Scaling Out With Apache Spark. DTL Meeting Slides based on

Apache Spark 11/10/15. Context. Reminder. Context. What is Spark? A GrowingStack

Apache Spark and Distributed Programming

From GWS to MapReduce: Google s Cloud Technology in the Early Days

Improving MapReduce Performance in Heterogeneous Environments

A REVIEW PAPER ON THE HADOOP DISTRIBUTED FILE SYSTEM

Unified Big Data Analytics Pipeline. 连 城

In-Memory BigData. Summer 2012, Technology Overview

Moving From Hadoop to Spark

Welcome to the unit of Hadoop Fundamentals on Hadoop architecture. I will begin with a terminology review and then cover the major components

What s next for the Berkeley Data Analytics Stack?

Real Time Data Processing using Spark Streaming

Architectural patterns for building real time applications with Apache HBase. Andrew Purtell Committer and PMC, Apache HBase

Spark and Shark: High-speed In-memory Analytics over Hadoop Data

BookKeeper. Flavio Junqueira Yahoo! Research, Barcelona. Hadoop in China 2011

Mambo Running Analytics on Enterprise Storage

Spark and the Big Data Library

In Memory Accelerator for MongoDB

Lambda Architecture. Near Real-Time Big Data Analytics Using Hadoop. January Website:

Oracle Big Data SQL Technical Update

How To Create A Data Visualization With Apache Spark And Zeppelin

The Flink Big Data Analytics Platform. Marton Balassi, Gyula Fora" {mbalassi,

Big Data Processing. Patrick Wendell Databricks

Processing Large Amounts of Images on Hadoop with OpenCV

Hadoop Parallel Data Processing

Hadoop s Entry into the Traditional Analytical DBMS Market. Daniel Abadi Yale University August 3 rd, 2010

BENCHMARKING CLOUD DATABASES CASE STUDY on HBASE, HADOOP and CASSANDRA USING YCSB

Brave New World: Hadoop vs. Spark

Spark Application on AWS EC2 Cluster:

Clash of the Titans: MapReduce vs. Spark for Large Scale Data Analytics

Big Data and Scripting Systems beyond Hadoop

Processing NGS Data with Hadoop-BAM and SeqPig

Extending Hadoop beyond MapReduce

I/O Considerations in Big Data Analytics

<Insert Picture Here> Big Data

Fast Data in the Era of Big Data: Tiwtter s Real-Time Related Query Suggestion Architecture

Big Data Course Highlights

Resilient Distributed Datasets: A Fault-Tolerant Abstraction for In-Memory Cluster Computing

Application Development. A Paradigm Shift

Big Fast Data Hadoop acceleration with Flash. June 2013

Spark: Cluster Computing with Working Sets

Hadoop Ecosystem Overview. CMSC 491 Hadoop-Based Distributed Computing Spring 2015 Adam Shook

Hadoop IST 734 SS CHUNG

Implementation of Spark Cluster Technique with SCALA

Complete Java Classes Hadoop Syllabus Contact No:

Hadoop Distributed File System. T Seminar On Multimedia Eero Kurkela

Big Data and Scripting Systems build on top of Hadoop

Apache Ignite TM (Incubating) - In- Memory Data Fabric Fast Data Meets Open Source

America s Most Wanted a metric to detect persistently faulty machines in Hadoop

Parallel Computing. Benson Muite. benson.

The Berkeley Data Analytics Stack: Present and Future

Petabyte Scale Data at Facebook. Dhruba Borthakur, Engineer at Facebook, SIGMOD, New York, June 2013

Architectures for massive data management

VariantSpark: Applying Spark-based machine learning methods to genomic information

Developing Scalable Java Applications with Cacheonix

Petabyte Scale Data at Facebook. Dhruba Borthakur, Engineer at Facebook, UC Berkeley, Nov 2012

BlobSeer: Towards efficient data storage management on large-scale, distributed systems

EXPERIMENTATION. HARRISON CARRANZA School of Computer Science and Mathematics

Shark: SQL and Rich Analytics at Scale

Fast Data in the Era of Big Data: Twitter s Real-

Data Warehousing and Analytics Infrastructure at Facebook. Ashish Thusoo & Dhruba Borthakur athusoo,dhruba@facebook.com

Big Data Research in the AMPLab: BDAS and Beyond

Cloudera Enterprise Reference Architecture for Google Cloud Platform Deployments

CDH AND BUSINESS CONTINUITY:

Processing of massive data: MapReduce. 2. Hadoop. New Trends In Distributed Systems MSc Software and Systems

Snapshots in Hadoop Distributed File System

RCFile: A Fast and Space-efficient Data Placement Structure in MapReduce-based Warehouse Systems CLOUD COMPUTING GROUP - LITAO DENG

Introduction to Hbase Gkavresis Giorgos 1470

Apache Ignite TM (Incubating) - In- Memory Data Fabric Fast Data Meets Open Source

Workshop on Hadoop with Big Data

Configuring Apache Derby for Performance and Durability Olav Sandstå

Transcription:

Tachyon: Reliable File Sharing at Memory- Speed Across Cluster Frameworks Haoyuan Li UC Berkeley

Outline Motivation System Design Evaluation Results Release Status Future Directions Outline Motivation Design Results Status Future

Memory is King Outline Motivation Design Results Status Future

Memory Trend RAM throughput increasing exponentially Outline Motivation Design Results Status Future

Disk Trend Disk throughput increasing slowly Outline Motivation Design Results Status Future

Consequence Memory locality key to achieve Interactive queries Fast query response Outline Motivation Design Results Status Future

Current Big Data Eco-system Many frameworks already leverage memory e.g. Spark, Shark, and other projects File sharing among jobs replicated to disk Replication enables fault-tolerance Problems Disk scan is slow for read. Synchronous disk replication for write is even slower. Outline Motivation Design Results Status Future

Tachyon Project Reliable file sharing at memory-speed across cluster frameworks/jobs Challenge How to achieve reliable file sharing without replication? Outline Motivation Design Results Status Future

Idea Re-computation (Lineage) based storage using memory aggressively. 1. One copy of data in memory (Fast) 2. Upon failure, re-compute data using lineage (Fault tolerant) Outline Motivation Design Results Status Future

Stack Outline Motivation Design Results Status Future

System Architecture Outline Motivation Design Results Status Future

Lineage Outline Motivation Design Results Status Future

Lineage Information Binary program Configuration Input Files List Output Files List Dependency Type Outline Motivation Design Results Status Future

Fault Recovery Time Re-computation Cost? Outline Motivation Design Results Status Future

Example Outline Motivation Design Results Status Future

Asynchronous Checkpoint 1. Better than using existing solutions even under failure. 2. Bounded recovery time (Naïve and Snapshot asynchronous checkpointing). Outline Motivation Design Results Status Future

Master Fault Tolerance Multiple masters Use ZooKeeper to elect a leader After crash workers contact new leader Update the state of leader with contents of caches Outline Motivation Design Results Status Future

Implementation Details 15,000+ lines of JAVA Thrift for data transport Underlayer file system supports HDFS, S3, localfs, GlusterFS Maven, Jenkins Outline Motivation Design Results Status Future

Sequential Read using Spark Flat Datacenter Storage Theoretical Maximum Disk Throughput Outline Motivation Design Results Status Future

Sequential Write using Spark Flat Datacenter Storage Theoretical Maximum Disk Throughput Outline Motivation Design Results Status Future

Realistic Workflow using Spark Outline Motivation Design Results Status Future

Realistic Workflow Under Failure Outline Motivation Design Results Status Future

Conviva Spark Query (I/O intensive) More than 75x speedup Tachyon outperforms Spark cache because of JAVA GC Outline Motivation Design Results Status Future

Conviva Spark Query (less I/O intensive) 12x speedup GC kicks in earlier for Spark cache Outline Motivation Design Results Status Future

Alpha Status Releases Developer Preview: V0.2.1 (4/25/2013) Contributions from: Outline Motivation Design Results Status Future

Alpha Status First read of files cached in-memory Writes go synchronously to HDFS (No lineage information in Developer Preview release) MapReduce and Spark can run without any code change (ser/de becomes the new bottleneck) Outline Motivation Design Results Status Future

Current Features Java-like file API Compatible with Hadoop Master fault tolerance Native support for raw tables WhiteList, PinList Command line interaction Web user interface Outline Motivation Design Results Status Future

Spark without Tachyon val file = sc.textfile( hdfs://ip:port/path ) Outline Motivation Design Results Status Future

Spark with Tachyon val file = sc.textfile( tachyon:// ip:port/path ) Outline Motivation Design Results Status Future

Shark without Tachyon CREATE TABLE orders_cached AS SELECT * FROM orders; Outline Motivation Design Results Status Future

Shark with Tachyon CREATE TABLE orders_tachyon AS SELECT * FROM orders; Outline Motivation Design Results Status Future

Experiments on Shark Shark (from 0.7) can store tables in Tachyon with fast columnar Ser/De 20 GB data / 5 machines Spark Cache Tachyon Table Full Scan 1.4 sec 1.5 sec GroupBys (10 GB Shark Memory) 50 90 sec 45 50 sec GroupBys (15 GB Shark Memory) 44 48 sec 37 45 sec Outline Motivation Design Results Status Future

Experiments on Shark Shark (from 0.7) can store tables in Tachyon with fast columnar Ser/De 20 GB data / 5 machines Spark Cache Tachyon Table Full Scan 1.4 sec 1.5 sec GroupBys (10 GB Shark Memory) 50 90 sec 45 50 sec GroupBys (15 GB Shark Memory) 44 48 sec 37 45 sec 4 * 100 GB TPC-H data / 17 machines Spark Cache Tachyon TPC-H Q1 65.68 sec 24.75 sec TPC-H Q2 438.49 sec 139.25 sec TPC-H Q3 467.79 sec 55.99 sec TPC-H Q4 457.50 sec 111.65 sec Outline Motivation Design Results Status Future

Future Efficient Ser/De support Fair sharing for memory Full support for lineage Next release is coming soon Outline Motivation Design Results Status Future

Acknowledgment Research Team: Haoyuan Li, Ali Ghodsi, Matei Zaharia, Eric Baldeschwieler, Scott Shenker, Ion Stoica Code Contributors: Haoyuan Li, Calvin Jia, Bill Zhao, Mark Hamstra, Rong Gu, Hobin Yoon, Vamsi Chitters, Reynold Xin, Srinivas Parayya, Dilip Joseph Outline Motivation Design Results Status Future

Questions? http://tachyon-project.org https://github.com/amplab/tachyon